Meta has released Muse Spark 1.1, a new in-house AI model it says is capable of competing on software coding tasks, and has opened it up to developers through a new Meta Model API. The move matters because it marks Meta's push to turn its heavy AI spending into a product that can rival offerings from OpenAI, Google and Anthropic, following its return to the AI race with the first Muse Spark model in April.

Meta describes the 1.1 version as a "step-change" over its predecessor, citing improvements shaped by developer feedback, including more advanced coding and complex bug-fixing, better support for end-to-end agentic and multi-agent workflows, and native multimodal understanding of images, videos and documents. The model is available now in Thinking mode via the Meta AI app and website, and through the Meta Model API in public preview for US developers, with $20 of free credits per new account. The launch follows Meta's controversial Muse Image generator and its broader effort to justify billions spent after high-profile hires and a company restructuring.